Seven Claudes Tried to Reject This Resume. Then It Rebuilt It.

The third of three posts on the English resume skills we imported today. The first covers a research-first tailoring pipeline; the second covers an ATS auditor.

Most resume feedback has one voice. A model reads the document, lists strengths and weaknesses, suggests stronger verbs, and stops. hiring-panel takes the opposite approach: seven independent reviewers, no shared notes, then a rebuild.

It scores 8.4, with almost no spread across dimensions.

The protocol

The skill runs a fixed sequence:

1. Get the resume and the job. If no role is given, ask one question — what job are you applying for? — then continue. Never invent employers, titles, dates, numbers or skills.
2. The automated screen. One sub-agent plays the ATS. It checks readability, clarity of titles and dates, keyword coverage, contact details, length and filler. It answers PASS or REJECT with at most five quoted reasons.
3. The hiring manager. One sub-agent decides INTERVIEW or NO, names what decided it in the first ten seconds, lists the interview questions it would ask, and quotes the three weakest lines.
4. Five readers, alone. Five sub-agents run at once, each seeing only the resume and the job — no other reviewer's notes. Each speaks as a different reader: the hiring manager (can this person do the job on day one?), another applicant (where is this resume weaker than mine?), the CEO (what did this person change, and by how much?), the recruiter (thirty seconds, top third only), and the teammate (how do they work with others?). Each answers HIRE, MAYBE or PASS in under 120 words, quoting the worst line and naming two more problems.
5. Merge. All seven answers become one numbered list of mistakes, worst first, deduplicated, each with the quoted line and the reviewers who flagged it.
6. Rebuild. The resume is rewritten with the same facts and every numbered mistake fixed. Missing facts are marked [ADD: ...]. The automated screen runs again on the new version.
7. Write the result. Two files in a dated folder: PANEL.md with the verdicts and the mistake list, and RESUME-IMPROVED.md with the new resume.

Why independence is the feature

The value is not that seven models are wiser than one. It is that they cannot anchor on each other.

A single reviewer tends to produce a balanced list: three strengths, three weaknesses, a polite conclusion. A panel of independent readers produces specific objections from incompatible incentives. The rival applicant compares your resume to an imagined competitor. The CEO asks what changed and by how much. The recruiter reads only the top third. Those are different questions, and a resume can pass one while failing another.

The quoted-line requirement makes the feedback actionable. "Strengthen your impact statements" is advice. "'Managed the roadmap' — what changed, and by how much?" is a defect report.

What the evaluator flagged

Two weaknesses kept the score at 8.4 instead of higher:

• The trigger has no when-not boundary. The description says to use it when the user types /hiring-panel with a resume, but never says when another skill is the better choice. It is a narrow, explicit invocation, which helps, but it still lacks the negative space we reward.
• The hiring manager appears twice. Step 3 runs a hiring-manager sub-agent; step 4's first reader is also "the hiring manager." The evaluator called this a conceptual overlap that slightly weakens clarity. It is not fatal — one is a verdict, the other is one of five independent lenses — but it is the kind of duplicate role that should be renamed or merged.

The rest is solid: a clear merge/rebuild/verify loop, safe file-write boundaries, and a final screen comparison that closes the loop instead of ending with advice.

The cost is real

Eight sub-agent calls per resume is not free. This is slower and more expensive than a single-pass review, and it is not the right tool for a quick sanity check.

The comparison with review-resume is useful. That skill scores 8.1 with a product-management checklist and one pass. Its trigger is 6.0 and its usability 7.2. hiring-panel scores 8.4 with no domain specialization at all. The difference is not knowledge; it is structure — independent judgments, a merge step, and a rebuild loop.

⚡ review-resume 8.1 Comprehensive PM resume review and tailoring against 10 best practices including XYZ+S formula, k… ⬇ 7

Run the panel when a resume is about to go out and the cost of a bad read is high. Run the cheaper checklist when you are still iterating.

A structured second opinion, not a prediction

Seven language models role-playing recruiters are not seven recruiters. The skill says so explicitly: it is a structured second opinion, not a guarantee of an interview or a job.

That honesty is part of why it works. The panel does not promise to predict hiring outcomes; it promises to surface objections you can act on. For a document whose whole job is to survive a ten-second read, that is the more useful promise.

The takeaway

Adversarial review is a reusable pattern beyond resumes: assign independent roles, hide their notes from each other, require quoted evidence, merge the objections, then rebuild and re-test. The skill's 8.4 is less about resume expertise than about encoding that loop so an agent cannot skip it.